The intuition behind regression is that we want to minimize the error in our prediction. To do so, OLS calculates the linear regression line that minimizes the (squared) residuals. 

Looking at the figures on the left, OLS minimizes the distances between the observations and the regression line to produce fewer residuals.